Model predictive control of a soft elbow exosuit reduces interaction torque

Nicholas Tacca, John Nassour, Gordon Cheng

Abstract: Incorporating soft materials into active exosuits has shown promise to provide assistance in a comfortable user-centric interface for the wearer. In order for individuals to be able to seamlessly operate an exosuit, user intention must be decoded so that the exosuit can move according to user expectations. Existing exosuit control methods aim to reduce an interaction torque between the wearer and exosuit inferred from the decoded intent from a high-level control scheme. Due to the nonlinearity response of the soft materials, however, sophisticated methods of optimal control from robotics have not been used to minimize this interaction torque. A model predictive controller (MPC) may be able to predict future interaction conflicts between the wearer and exosuit assistance in order to preemptively reduce this interaction torque to provide assistance in line with intentions. Here, we experimentally approximate a model for our soft pneumatic elbow exosuit and demonstrate the feasibility of using a low-level MPC to reduce interaction torque determined from a gravity compensation high-level control approach. We demonstrate that in a step flexion response of the system, the MPC reduces oscillations around the target exosuit torque compared to on/off and PID controllers. These results demonstrate that using a model-based predictive approach reduces the interaction between the wearer and exosuit for a more naturalistic interface.